Kimio Tanaka seems to be a phtographer and not too much into technical details. There have been several DSLR released last year (e.g. 90D, D6, 1D X iii, D780) and I expect there will be some more coming this year D850 successor, D500 successor? Also the tracking is available in todays Pentax cameras already.
The deep learning mode today is stated (for KP and K-1) as available in green mode only.
The use of the RGB-Sensor together with the prime processor to make the Analysis System was stated in the K-3 i.e. 2013 already (
Feature 1?K-3 | RICOH IMAGING) - shape and color detection, subject detection. The 1st mention for Nikon consumer cameras to my knowledge is the D90 in 2008 (
https://org.imaging.nikon.com/lineup/dslr/d90/pdf/d90_16p.pdf).
There is a lot of discussion regarding Pentax AF capability. I don't think is a Pentax specific issue, it is a prize range issue. The question is about speed of a camera to allow the control system to work e.g. mirror black out times, processing time, buffer read-out time, shutter, .... Todays Pentax cameras are rather slow in that respect. With the K-3 iii that should change.
On the more costly camera side, in Nikon and Canon cameras (e.g. D500) you can give the AF control system a target by setting in the menue "face detect" for example. To what was posted earlier for the K-3 iii that feature will be available. It would be new for Pentax, also the speed of the camera is on a different level.
Looking forward to see the result with respect to AF precision.
---------- Post added 01-07-21 at 06:07 PM ----------
In industry, deep learning or artificial intelligence can be a name of analysis of a very large number of similar situations available from the internet and using that information to help a control system to maintain the target. Meant is really analysis of similar situations and programming that into the control system. E.g. a person is in camera focus, both eyes visible, then turns it head, only one eye visible, which is more difficult to identify as a face. Give the camera a few sample images and it would know from the initial start with both eyes vsibile how the image structure evolves over time. Camera could now swith eye and maintain focus.
What is usually not meant is, that the camera learns from the photographer using that specific camera in a certain way.
The K-1 has feature pages, which explain the deep learning feature.